victor-rag

RAG (Retrieval-Augmented Generation) Vertical for Victor AI

A complete RAG implementation showcasing document ingestion, vector search, and knowledge management with Victor AI.

Features

• 📄 Multi-format Document Ingestion

  • PDF, Markdown, Text, Code files
  • Semantic chunking with configurable overlap
  • Automatic document type detection

• 🔍 Hybrid Search

  • Vector search (semantic similarity)
  • Full-text search (keyword matching)
  • Combined reranking for best results

• 💾 Embedded Vector Storage

  • LanceDB (no server required)
  • Persistent local storage
  • Fast similarity search

• 🎯 Query Enhancement

  • Context retrieval from relevant documents
  • Source attribution and citations
  • Confidence scoring

Installation

# Install with Victor core
pip install victor-ai

# Install RAG vertical
pip install victor-rag

Quick Start

from victor.framework import Agent

# Create agent with RAG vertical
agent = await Agent.create(
    provider="ollama",
    model="qwen2.5-coder:7b",
    vertical="rag"
)

# Ingest a document
await agent.chat("Ingest README.md into the knowledge base")

# Query the knowledge base
result = await agent.chat("What does this project do?")

Available Tools

Once installed, the RAG vertical provides these tools:

• rag_ingest - Ingest documents into the knowledge base
• rag_search - Search for relevant document chunks
• rag_query - Query with context retrieval
• rag_list - List all indexed documents
• rag_delete - Delete documents from knowledge base
• rag_stats - Get knowledge base statistics

System Prompt

The RAG vertical includes specialized prompt contributions:

• Query Strategy: When to use semantic vs keyword search
• Citation Format: How to reference sources
• Document Handling: Chunking strategies by document type
• Quality Rules: Grounding rules to prevent hallucination

Configuration

The RAG vertical can be configured via environment variables:

# Vector store location
export VICTOR_RAG_DB_PATH=./rag_db

# Chunking configuration
export VICTOR_RAG_CHUNK_SIZE=512
export VICTOR_RAG_CHUNK_OVERLAP=50

# Search configuration
export VICTOR_RAG_TOP_K=5
export VICTOR_RAG_MIN_CONFIDENCE=0.3

Development

# Install with dev dependencies
pip install -e ".[dev]"

# Run tests
pytest tests/

# Format code
black victor_rag/

# Type check
mypy victor_rag/

License

Apache License 2.0 - see LICENSE for details.

Links

• Victor AI: https://github.com/anvai-labs/victor
• Documentation: https://docs.victor.dev/verticals/rag
• Victor Registry: https://github.com/anvai-labs/victor-registry